Occludra

AI Security Gateway (AISG) sits between your app and any LLM provider — scanning every request for PII, injection attacks, and agent retry loops before anything reaches the model. Security: • Auto-redacts 30+ PII entity types in text + images (OCR vision DLP) • Blocks prompt injection (jailbreaks, DAN, instruction overrides) • Recursive loop protection — SHA-256 fingerprint detection kills agent retry loops (LangChain, CrewAI, AutoGPT). A single GPT-4 loop drains $108/hr undetected. • 4-tier RBAC — Owner, Admin, Member, Viewer with 17 granular permissions • SAML SSO — Okta, Azure AD, Google Workspace with auto-provisioning Real-Time Alerting: • HMAC-SHA256 signed webhook alerts for PII blocks, injection attempts, budget exhaustion, and loop detection • Push to Slack, PagerDuty, or any SIEM — no polling required • SIEM Connectors — stream events to Splunk HEC, Datadog Logs, or Microsoft Sentinel Compliance: • EU AI Act Article 12 logging (enforcement: Aug 2, 2026) • Hash-chained tamper-evident audit records — SHA-256 fingerprints only, no raw content stored • JSONL export + chain verification API • 10-year retention, GDPR/HIPAA/PCI-DSS ready Cost Control: • Smart routing across 8+ providers, 380+ models — cheapest auto-selected • Hard budget enforcement per project • BYOK with 0% markup • Per-project budget enforcement with hard spending caps Integration: • OpenAI SDK compatible — 2-line code change • Python SDK (pip install aisg) — typed exceptions for loop, budget, PII errors • Streaming (SSE) fully supported • Stateless processing — zero content retention Deploy: • Hybrid VPC — compiled Go proxy in your network, cloud dashboard for policies • Self-host via Docker (Apache 2.0) • Managed cloud. 1M free credits, no credit card. github.com/aisecuritygateway/aisecuritygateway